DIGITAL LIFE

OpenAI blocks Russian-origin AI influence campaign

An AI-driven influence campaign used ChatGPT to create posts promoting an entity presented as an Israel-based think tank. OpenAI blocked the associated accounts after concluding they likely originated in Russia.

The operation promoted the "International Burke Institute" through content published on X, LinkedIn, Facebook, Substack, and Telegram. The institute featured experts, academic articles, and a "sovereignty index," but an analysis revealed plagiarized texts, false attributions, and content favorable to Russia.

Key points:

OpenAI blocked a set of ChatGPT accounts likely originating in Russia.

The operators used the platform to create posts and comments for social media.

Analysis found 34 plagiarized articles in a sample of 36 pieces of content associated with the institute.

Most posts saw little engagement, although some Telegram channels amassed between 10,000 and 20,000 followers.

OpenAI announced the account blocks on August 25 in a report on the influence operation. The company stated that users accessed ChatGPT from Russia using virtual private networks (VPNs).

Access to OpenAI models is not available in Russia. VPNs allowed the accounts to mask their actual location and bypass this restriction.

The operators wrote prompts in Russian but requested content primarily in English. Some requests specified the removal of linguistic traits that might reveal the authors' Russian origin.

OpenAI identified ChatGPT-generated text in posts on X, LinkedIn, Facebook, Substack, and Telegram. Some accounts used the International Burke Institute's name and visual identity. Others posed as ordinary users and primarily shared links to the entity's content.

ChatGPT was also used to generate replies to posts by real users on Substack. These messages typically recommended that readers follow the institute's channel. The investigation also uncovered German-language posts targeting Ukraine, the European Union, and the German government. Another operator requested images for approximately 12 Telegram channels focused on Germany, the United States, France, Poland, and Turkey.

Attribution requires caution. OpenAI considers it highly likely that the accounts originated in Russia but presented no evidence of a direct link to the Russian government.

The institute used plagiarized academic content...Most of the activity promoted the International Burke Institute, also known by the acronym IBI. Its website was registered in February 2025 and presented the organization as an international community of experts based in Israel.

The page linked well-known figures from academia and politics to the institute. Among them were Francis Fukuyama, Noam Chomsky, and former U.S. Secretary of State Mike Pompeo.

OpenAI’s research analyzed a sample of 36 articles published between September 2025 and May 2026. The company concluded that 34 of them had been copied from other online sources.

Some texts were years old. Others were attributed to authors who had not written them. One article about the China-Pakistan economic corridor, for instance, reproduced a work published by Cambridge University Press but attributed the content to a different researcher.

In another instance, the site linked an article on migration policy to an Australian food chemistry professor. The original text had been published by the Migration Policy Institute and was authored by others.

An independent investigation by *Le Monde* also found references to alleged partnerships with international organizations. UNICEF confirmed to the newspaper that the International Burke Institute was not an approved partner and that there was no record of any joint work.

These elements created an appearance of academic authority. The plagiarized content gave the site a documentary foundation that could seem legitimate to a reader who did not verify the original authors and publications. OpenAI found no evidence that the items hosted on the site had been produced by its models. ChatGPT's role focused primarily on creating promotional posts for social media.

Artificial intelligence supported the distribution of the campaign... The International Burke Institute also published an index designed to assess national sovereignty. The methodology placed Russia in a favorable position while presenting negative assessments of France, Germany, the European Union, and the United States.

The texts argued that Western European countries had lost political and economic autonomy. They also criticized European support for Ukraine and advocated for better relations with Russia.

Using a proprietary index lent the messages a quantitative appearance. Operators could present political opinions as the results of an institutional methodology, even without demonstrating the validity of the criteria used.

ChatGPT served to amplify the distribution of this content. Instead of building an entire infrastructure from scratch, the model generated messages tailored to different platforms and audiences.

This distinction is crucial for understanding the role of artificial intelligence in influence operations. While the technology can reduce the time and cost required to produce content in multiple languages, it still relies on websites, accounts, channels, and distribution mechanisms.

In this instance, the use of ChatGPT also contributed to the operation's detection. By analyzing accounts and requests, OpenAI was able to link scattered posts to the International Burke Institute and the channels promoting its content.

OpenAI considers the campaign's immediate impact to have been limited. Social media posts generally garnered few views and interactions, and the institute's official accounts had few subscribers.

Telegram channels boasted larger audiences; according to the company, several had between 10,000 and 20,000 followers. These figures do not, in themselves, confirm that the followers were authentic or that they engaged with the content.

Using the impact scale developed by the Brookings Institution, OpenAI placed the operation at the lower end of category three. This classification corresponds to activity spanning multiple platforms, with some evidence of contact with real audiences.

The case is significant primarily due to its infrastructure. The operation brought together an entity with an institutional appearance, alleged experts, academic content, a proprietary index, and distribution channels across several countries.

*Le Monde* also identified individuals physically present in Israel who were associated with the institute or the promotion of the index. OpenAI stated that it was unable to determine the relationship between these individuals, the IBI, and the operators based in Russia.

To date, there is also no proven link between the institute and the Social Design Agency, an entity associated with other Russian disinformation campaigns. However, previously released documents outlined plans to establish a think tank in Israel capable of distributing content through Western media outlets.

The case demonstrates that artificial intelligence can serve as a component of broader operations aimed at manufacturing credibility. A limited initial reach does not eliminate the risk that such an infrastructure could expand, gain recognition, and serve as an apparent source for political content.

mundophone